tfm.core.actions.RecoveryAction

Train action to recover from loss blowup.

Checks the loss value by the given threshold. If applicable, recover the model by reading the checkpoint on disk.

Methods

__call__

View source

Recovers the training by triggering checkpoint restoration.